coefplot of results of `lincom`

[Tom Ford]

Hello,

I am using lincom to aggregate the results of my difference in difference model using the estimator from Sun and Abraham (2020).

I would like to do a coefplot of the estimated aggregated estimate from lincom. Does anyone konw how I can do it?


take the following example.

eventstudyinteract dv1 $eventtt182 if anytrai==1, vce(cluster prod) 
> absorb(prod yr rsph audcmp audseq auditann) cohort(yr_first_post_trai182) control
> _cohort(lastcohort182)
(obs=4,883)

IW estimates for dynamic effects                        Number of obs =  5,913
Absorbing 6 HDFE groups                                 F(21, 2228)   =   2.09
                                                        Prob > F      = 0.0026
                                                        R-squared     = 0.6140
                                                        Adj R-squared = 0.2464
                                                        Root MSE      = 0.3800
                               (Std. err. adjusted for 2,229 clusters in prod)
------------------------------------------------------------------------------
             |               Robust
auditratin~s | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
    L0ttt182 |   .1218806    .040001     3.05   0.002     .0434374    .2003238
    L1ttt182 |   .2393228   .0596118     4.01   0.000     .1224224    .3562232
    L2ttt182 |   .1201224   .0668075     1.80   0.072    -.0108891    .2511339
    F2ttt182 |   .0404785   .0425113     0.95   0.341    -.0428875    .1238445
    F3ttt182 |   .0675974   .0537379     1.26   0.209    -.0377842     .172979
    F4ttt182 |   .0120441   .0752037     0.16   0.873    -.1354326    .1595208
------------------------------------------------------------------------------

* Following the .shape file of `eventstudyinteract` I use the `lincom` command to aggregate event study estimates. 

matrix b = e(b_iw)

. matrix V = e(V_iw)

. ereturn post b V

. lincom (L0ttt182 + L1ttt182 + L2ttt182)/3

 ( 1)  .3333333*L0ttt182 + .3333333*L1ttt182 + .3333333*L2ttt182 = 0

------------------------------------------------------------------------------
             | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
         (1) |   .1604419   .0484821     3.31   0.001     .0654187    .2554652
------------------------------------------------------------------------------


* Now I run the same model on a different dependent variable


. eventstudyinteract dv2 $eventtt182 if anytrai==1, vce(cluster p
> rod) absorb(prod yr rsph audcmp audseq auditann) cohort(yr_first_post_trai182) co
> ntrol_cohort(lastcohort182)
(obs=4,883)

IW estimates for dynamic effects                        Number of obs =  5,913
Absorbing 6 HDFE groups                                 F(21, 2228)   =   1.10
                                                        Prob > F      = 0.3449
                                                        R-squared     = 0.5438
                                                        Adj R-squared = 0.1093
                                                        Root MSE      = 0.3491
                               (Std. err. adjusted for 2,229 clusters in prod)
------------------------------------------------------------------------------
             |               Robust
par~inv_pass | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
    L0ttt182 |   .0461567   .0343948     1.34   0.180    -.0212925    .1136059
    L1ttt182 |    .054493   .0552511     0.99   0.324     -.053856    .1628419
    L2ttt182 |   .0184704   .0557928     0.33   0.741     -.090941    .1278818
    F2ttt182 |   .0150241   .0393211     0.38   0.702    -.0620857    .0921339
    F3ttt182 |  -.0337918   .0490002    -0.69   0.491    -.1298825     .062299
    F4ttt182 |   -.006068    .066574    -0.09   0.927    -.1366216    .1244855
------------------------------------------------------------------------------

. 
end of do-file

. do "/var/folders/d8/4ly8xgqn22bc36_bmmy54625n0kgs3/T//SD55008.000000"

. */
. matrix b = e(b_iw)

. matrix V = e(V_iw)

. ereturn post b V

. lincom (L0ttt182 + L1ttt182 + L2ttt182)/3

 ( 1)  .3333333*L0ttt182 + .3333333*L1ttt182 + .3333333*L2ttt182 = 0

------------------------------------------------------------------------------
             | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
         (1) |   .0397067   .0428879     0.93   0.355    -.0443521    .1237655
------------------------------------------------------------------------------


* here I would like to create my coefplot with the impact of my aggregated treatment effect on the two different dependent variables.

In sum, I have my two regressions on different DVs and I do the lincom of post treatment effects. I would like to plot these results using coefplot. does anyone knows how to do it?


thanks a lot

[Jad Tamimi]

This is my exact problem right now lol, did u figure it out yet?

[Dimitriy V. Masterov]

Here's how I might do this. Note the use of a reproducible example that anyone come run. Providing one in your question, even if it errors out, makes it more likely that someone will answer since it requires less effort to tweak broken code than making a working example from scratch.

webuse nlswork, clear
gen union_year = year if union == 1
bysort idcode: egen first_union = min(union_year)
drop union_year
gen ry = year - first_union
gen never_union = (first_union == .)
forvalues k = 18(-1)2 {
    gen g_`k' = ry == -`k'
}
forvalues k = 0/18 {
    gen g`k' = ry == `k'
}
capture ssc install avar
capture ssc install xlincom
set matsize 800
eventstudyinteract ln_wage g_* g0-g18, cohort(first_union) control_cohort(never_union) covariates(south) absorb(i.idcode i.year) vce(cluster idcode)
matrix b = e(b_iw)
matrix V = e(V_iw)
ereturn post b V
xlincom [(g0 + g1 + g2 + g3 + g4)/5], post
estimates store b1
eventstudyinteract wks_work g_* g0-g18, cohort(first_union) control_cohort(never_union) covariates(south) absorb(i.idcode i.year) vce(cluster idcode)
matrix b = e(b_iw)
matrix V = e(V_iw)
ereturn post b V
xlincom [(g0 + g1 + g2 + g3 + g4)/5], post
estimates store b2
coefplot b1 b2